Understanding predictions of drug profiles using explainable machine learning models

Area of Science:

• Computational chemistry and cheminformatics
• Pharmacology and drug discovery
• Artificial intelligence in medicine

Background:

• Absorption, distribution, metabolism, and excretion (ADME) properties are critical determinants of drug efficacy and safety.
• Predicting ADME properties early in drug design accelerates the identification of viable drug candidates.
• Understanding the molecular basis of ADME is essential for optimizing drug performance.

Purpose of the Study:

• To predict ADME molecular properties using explainable Machine Learning (ML) models.
• To identify and quantify the impact of specific molecular features on ADME property predictions.
• To enhance drug design by elucidating the contribution of molecular characteristics to ADME behavior.

Main Methods:

• Utilizing explainable Machine Learning (ML) models for ADME property prediction.
• Employing feature permutation techniques to estimate the relative importance of molecular features.
• Applying SHAP (SHapley Additive exPlanations) values to measure the individual impact of features.

Main Results:

• Identification of specific molecular descriptors relevant to each ADME property.
• Quantification of the impact of these molecular descriptors on ADME property prediction accuracy.
• Demonstration of feature importance in predicting ADME outcomes.

Conclusions:

• Explainable ML models offer detailed insights into molecular feature contributions to ADME predictions.
• These models support drug candidate selection by clarifying the influence of molecular features.
• The study highlights the utility of interpretable AI in advancing pharmaceutical research.